Solutions to the 2012 CMI BSc Entrance Examination

Part A: 5 problems 6 marks. Part B: 7 out of 9 problems 10 marks.

A1. Find the number of real solutions to the equation x = 99 sin(x).

x
The number of solutions is the number of times the line y = 99 meets the graph of
y = sin(x). This can occur only for x [99, 99] because sin(x) has range [1, 1]. Also
sin(x) is periodic with period 2. For x 0, the two graphs meet twice in each cycle
of sin(x), both intersections occurring in the first half of the cycle. There are 50 such
half-cycles from x = 0 to x = 99, over intervals [0, 1], [2, 3], . . . , [98, 99]. So there are 100
non-negative solutions. Similarly there are 100 solutions 0 because both graphs are odd.
Since x = 0 is counted twice, the total number of solutions is 100 + 100 1 = 199.

A2. A differentiable function f : R R satisfies f (1) = 2, f (2) = 3 and f (3) = 1. Show

that f 0 (x) = 0 for some x.

Since f is differentiable, it is continuous. By the intermediate value theorem, there is

a (2, 3) with f (a) = 2 = f (1). Now by Rolles theorem there is x (1, a) with f 0 (x) = 0.
OR The continuous function f over the closed interval [1, 2] must attain its absolute
maximum, which cannot be at either endpoint (since f (2) > f (1) and f (2) > f (3)). So
the maximum must be at an interior point x and then f 0 (x) = 0. OR By the mean
value theorem, f 0 (y) = 1 > 0 for some y (1, 2) and f 0 (z) = 2 < 0 for some z (2, 3).
So f 0 (x) = 0 for some x (y, z) since for a differentiable f , the function f 0 satisfies the
intermediate value property by Darbouxs theorem. (This is important to say because f 0
need not be continuous.)

ln(12)
A3. Show that ln(18) is irrational.

ln(12)
ln(18)= log18 (12). Suppose this is rational, say = ab where a, b are integers with b 6= 0.
a
Then 18 b = 12, so 18a = 12b . By factoring into primes this gives 32a 2a = 3b 22b , which
by unique factorization can happen only if 2a = b and a = 2b. But this gives a = b = 0,
ln(2)
a contradiction. (Alternatively and similarly, prove that r = ln(3) is irrational and show
ln(12) ln 3+2 ln 2 1+2r
that rationality of ln(18) = 2 ln 3+ln a = 2+r would force r to be rational as well.)

A4. Show that

x100 ln(x)
lim = 0.
x ex tan1 ( 3 + sin x)

There is a positive constant c such that tan1 ( 3 +sin x) > c for any x, e.g. c = tan1 (0.04)
will work since > 3.12, sin(x) 1 and tan1 is an increasing function. Moreover
ln(x) < x for x > 0. So the given ratio is sandwiched between 0 and x101 /cex . Now use
LHospitals rule repeatedly.

1
A5. a) n identical chocolates are to be distributed among the k students in Tinkus class.
Find the probability that Tinku gets at least one chocolate, assuming that the n chocolates
are handed out one by one in n independent steps. At each step, one chocolate is given to
a randomly chosen student, with each student having equal chance to receive it.

P(Tinku gets at least one chocolate) = 1 P(Tinku gets none) = 1 (1 k1 )n , because in

each of the independent steps the probability of Tinku not getting a chocolate is 1 k1 .

b) Solve the same problem assuming instead that all distributions are equally likely. You
are given that the number of such distributions is n+k1


k1 . (Here all chocolates are con-
sidered interchangeable but students are considered different.)

There are (n1)+k1

k1 distributions in which Tinku gets at least a chocolate: give Tinku a
chocolate and then use the given formula to find number of distributions

n+k1
of then remaining
(n1)+k1
n 1 chocolates among k students. So the answer is k1 / k1 = n+k1 . OR
The number of distributions in which Tinku gets no chocolate = number of distributions
n+k2


of n chocolates among the remaining k 1 students = k2 . So the desired probability
is 1 n+k2

n+k1
n
k2 / k1 = n+k1 .

B1. a) Find a polynomial p(x) with real coefficients such that p( 2 + i) = 0.

Non-real
roots of a polynomial with
real coefficients occur in conjugate pairs. p(x) =
2
(x ( 2 + i))(x ( 2 i)) = x 2 2x + 3 works.

b) Find a polynomial
q(x) with rational coefficients and having the smallest possible degree
such that
q( 2 + i) = 0. Show that any other polynomial with rational coefficients and
having 2 + i as a root has q(x) as a factor.

2 + i satisfies x2 2 2x + 3 = 0, i.e., x2 + 3 = 2 2x and so satisfies (x2 + 3)2 =
8x2 . So q(x) = (x2 + 3)2 8x2 works. A cubic with rational coefficients will not work
because, after
dividing by the necessarily rational leading coefficient, it must be of the
2
form (x 2 2x + 3)(x r). This forces the coefficients 3r and 2 2 r to be both
rational, which is impossible.

Let f (x) be a polynomial with rational coefficients such that f ( 2 + i) = 0. Divide f (x)
by q(x) using long division to get quotient a(x) andremainder b(x), both polynomials
with rational coefficients.
Using f ( 2 + i) = 0 and q( 2 + i) = 0 in the equation f (x) =
q(x)a(x) + b(x) gives b( 2 + i) = 0. Now if the remainder b(x)is a nonzero polynomial,
then it would have rational coefficients, degree less than 4 and 2 + i as a root. But we
just proved that this is impossible. Hence b(x) = 0, i.e., f (x) is a multiple of q(x).

B2. a) Let E, F, G and H respectively be the midpoints of the sides AB, BC, CD and
DA of a convex quadrilateral ABCD. Show that EFGH is a parallelogram whose area is
half that of ABCD.

2
Consider the diagonals AC and BD. By the basic proportionality theorem in triangle ABC,
we get that EF and AC are parallel and AC = 2 EF. Moreover, ABC and EBF are similar.
Using triangles ADC and HDG, we similarly get that AC is parallel to HG, AC = 2 HG.
Thus EF and HG are parallel. Likewise FG and EH are parallel (both parallel to BD), so
EFGH is a parallelogram. Also by similarity, Area(ABC) = 4 Area(EBF), Area(ADC) =
4 Area(HDG), Area(BAD) = 4 Area(EAH) and Area(BCD) = 4 Area(FCG). (Note. So
far convexity of ABCD is unnecessary. But the next steps need it, draw pictures and see.)
Area(EFGH) = Area(ABCD) [Area(EBF) + Area(FCG) + Area(GDH) + Area(HAE)]
= Area(ABCD) 41 [Area(ABC)+ Area(BCD) + Area(CDA) + Area(DAB)]
= Area(ABCD) 12 Area(ABCD) = 12 Area(ABCD).

b) Let E = (0, 0), F = (0, 1), G = (1, 1), H = (1, 0). Find all points A = (p, q) in the
first quadrant such that E, F, G and H respectively are the midpoints of the sides AB,
BC, CD and DA of a convex quadrilateral ABCD.

If A = (p, q) is such a point, then E = (0,0) being the midpoint of AB is equivalent to

having B = (p, q). Similarly we get C = (p, q 2), D = (2 p, q). In particular AC =
BD = 2, AC is vertical and BD horizontal. By the reasoning in part a), these facts imply
that the quadrilateral constructed from the midpoints of the sides of ABCD is a square of
side 1. So we just need to ensure that the listed coordinates make ABCD into a convex
quadrilateral. This happens if and only if p, q are both positive (which is given) and < 1.
It is easy to see that these conditions are sufficient to make ABCD a convex quadrilateral.
For necessity see the following (pictures will help). If p > 1 then A will be to the right of
H and so D to the left of H. If q > 1, then B will be below F and so C will be above F. If
p or q = 1, then three of the points A, B, C, D become collinear. In all cases ABCD will
not be a convex quadrilateral. If both p, q > 1, ABCD will even be self-intersecting.

B3. a) We want to choose subsets A1 , A2 , . . . , Ak of {1, 2, . . . , n} such that any two of the
chosen subsets have nonempty intersection. Show that the size k of any such collection of
subsets is at most 2n1 .

If a set A is in such a collection C, then the complement of A cannot be in C. Therefore

|C| 21 (total number of subsets of {1, 2, . . . , n}) = 21 2n = 2n1 .

b) For n > 2 show that we can always find a collection of 2n1 subsets A1 , A2 , . . . of
{1, 2, . . . , n} such that any two of the Ai intersect, but the intersection of all Ai is empty.

There are many ways to build such a collection, e.g., take all 2n1 subsets of {1, 2, . . . , n}
containing 1, remove the singleton set {1} and instead include its complement. OR
Note that for n = 3, the four sets {1, 2}, {2, 3}, {1, 3}, {1, 2, 3} give a (unique) solution.
For n > 3 take the union of each of these 4 sets with all 2n3 subsets of {4, . . . , n}. OR
For n = 2k +
1, take
all subsets of {1, 2, . . . , n} of size > k. Any two of these will intersect.
Now use ni = ni n
. For n = 2k, take all subsets of size > k along with half the subsets
of size k, namely those containing a fixed number. (Check the details.)

3
B4. Define
10 9
X 1 1 X 1 1
x= and y= .
i=1
10 3 1 + ( 10i3 )2 i=0
10 3 1 + ( 10i3 )2
x+y
Show that a) x < 6 < y and b) 2 < 6 . (Hint: Relate these sums to an integral.)

a) Let f (t) = 1/(1 + t2 ). Then y and x are respectively the left and right hand
Riemann
1
sums for f over the interval [0, 3 ] using 10 equal parts, each of width 1/10 3. Since f is
a positive decreasing function, y overestimates the area under f over the given interval and
1
R 1/3 1

1/ 3
x underestimates it. The area under f over [0, 3 ] is 0 f (t)dt = tan (t)|0 = 6 , so

x < 6 < y. Note. Different normalizations are possible for f , e.g., the more simpleminded
choice f (t) = 1013 1+( 1t )2 considered over the interval [0,10] will work too.
10 3

b) x+y 2 can be interpreted as the sum of areas of 10 trapezoids as follows. Dividing

[0, 13 ] into 10 equal parts, let the i-th subinterval be [ti1 , ti ] with i = 0, 1, . . . , 10. Then
the i-th trapezoid has base [ti1 , ti ] and it has two vertical sides, the left one of height
f (ti1 ) and the right one of height f (ti ) (draw a picture and see). So we have to prove
that the total area of trapezoids is less than the area under f . For this we should check
concavity of f (draw pictures and see why). Check that over the interval (0, 13 ), we have
2
6t 2
f 00 (t) = (1+t2 )3 < 0, so f is concave down and hence each trapezoid lies completely below

the graph of f .

B5. Using the steps below, find the value of x2012 + x2012 , where x + x1 = 5+1
2 .

a) For any real r, show that |r + r1 | 2. What does this tell you about the given x?

Because of the absolute value we may assume that rp> 0 by replacing r with r if
necessary.
1
Now use AM-GM inequality or the fact that ( r 1/r) 0. Since x+x = 5+1
2
2 < 2,
given x must be a non-real (complex) number.

b) Show that cos( 5 ) = 5+1
4 , e.g. compare sin( 2 3
5 ) and sin( 5 ).

Let = 5 . Then sin(2) = sin( 2) = sin(3). Using the formulas for sin(2) and
sin(3), canceling sin (it is nonzero) and substituting sin2 = 1 cos2, gives the
quadratic equation 4 cos2 2 cos 1 = 0. Since cos > 0, we get cos = 5+1
4 .

c) Combine conclusions of parts a and b to express x and therefore the desired quantity
in a suitable form.

Let x = dei = d(cos + isin ). Then x1 = d1 ei = d1 (cos i sin ). Adding

i
and using that x + x1 = 5+1 2 = 2 cos( 5 ), we get d = 1 and = . So x = e 5 and

x2012 + x2012 = 2 cos( 2012
5 ) = 2 cos(402 + 2
5 ) = 2 cos( 2
5 ) = 2 cos 2
( 5 ) 1 = 51
2 .

4
B6. For n > 1, a configuration consists of 2n distinct points in a plane, n of them red, the
remaining n blue, with no three points collinear. A pairing consists of n line segments, each
with one blue and one red endpoint, such that each of the given 2n points is an endpoint
of exactly one segment. Prove the following.
a) For any configuration, there is a pairing in which no two of the n segments intersect.
(Hint: consider total length of segments.)

For any configuration, there are only finitely many pairings. Choose one with least possible
total length of segments. Here no two of the n segments can interest, because if RB and
R0 B 0 intersect in point X then we get a contradiction as follows. Using triangle inequality
in triangles RXB 0 and R0 XB, we get RB 0 + R0 B < RB + R0 B 0 (draw a picture). So
replacing RB and R0 B 0 with R0 B and RB 0 would give a pairing with smaller total length.

b) Given n red points (no three collinear), we can place n blue points such that any pairing
in the resulting configuration will have two segments that do not intersect. (Hint: First
consider the case n = 2.)

For n = 2, place the two blue points on opposite sides of the line passing through the given
two red points. There are two possible pairings and the two segments in either one do not
intersect. We use a similar idea in general. Given n red points, find a triangle ABC such
that A is a red point and all other red points are inside triangle ABC. (This is always
possible. Why?) Place one blue point at B and all other blue points in the region opposite
to triangle ABC at vertex C. (More precisely, let C be between A and A0 and also between
B and B 0 . Place the remaining blue points inside triangle A0 CB 0 .) Now in any pairing, if
A and B are connected, then AB will not intersect any other segment. Otherwise the two
segments having A and B as vertices will not intersect. Draw a picture to see this.

B7. A sequence of integers cn starts with c0 = 0 and satisfies cn+2 = acn+1 + bcn for
n 0, where a and b are integers. For any positive integer k with gcd(k, b) = 1, show that
cn is divisible by k for infinitely many n.

Consider pairs of consecutive entries of the sequence modulo k, i.e., (cn , cn+1 ), where a
denotes a modulo k. Since there are only finitely many possibilities (namely k 2 ), some pair
of consecutive residues will repeat. Suppose (ci , ci+1 ) = (ci+p , ci+p+1 ) for some i. We will
show that in fact the previous equation holds for all i, i.e., whole sequence of consecutive
pairs is periodic. This will prove in particular that (c0 , c1 ) = (cp , cp+1 ) = (c2p , c2p+1 ) = .
Since c0 = 0 is divisible by k, so is cip for all i.
The equation cn+2 = acn+1 + bcn shows that bcn = cn+2 acn+1 . Now gcd(k, b) = 1 means
b is invertible modulo k, i.e., there is a b0 with bb0 = 1. Therefore cn = b0 (cn+2 acn+1 ).
Thus knowing a pair of consecutive residues uniquely determines the previous residue
(this is why we considered pairs of residues). Therefore (ci , ci+1 ) = (ci+p , ci+p+1 ) implies
(ci1 , ci ) = (ci+p1 , ci+p ) and (by the given recurrence) (ci+1 , ci+2 ) = (ci+p+1 , ci+p+2 ).
Thus the whole sequence (cn , cn+1 ) becomes periodic as soon as a single such pair repeats.

5
B8. Let f (x) be a polynomial with integer coefficients such that for each nonnegative
integer n, f (n) = a perfect power of a prime number, i.e., of the form pk , where p is prime
and k a positive integer. (p and k can vary with n.) Show that f must be a constant
polynomial using the following steps or otherwise.
a) If such a polynomial f (x) exists, then there is a polynomial g(x) with integer coefficients
such that for each nonnegative integer n, g(n) = a perfect power of a fixed prime number.

Write f (x) = an xn + an1 xn1 + + a1 x + a0 . Then a0 = f (0) = pk for some prime

p and integer k > 0. Define g(x) = f (px). Then g(x) is a polynomial such that for each
nonnegative integer n, g(n) = f (pn) = a perfect power of a prime number. This prime
number has to be p, because by evaluating we see that g(n) = f (pn) is divisible by p.

b) Show that a polynomial g(x) as in part a must be constant.

Let g(x) = bn xn + bn1 xn1 + + b1 x + b0 . Then b0 = g(0) = pk . Consider g(mpk+1 ) =

bn (mpk+1 )n +bn1 (mpk+1 )n1 + +b1 (mpk+1 )+pk . Clearly for each non-negative integer
m, this expression is divisible by pk , but not by pk+1 (since it is pk modulo pk+1 ). This
forces g(mpk+1 ) = pk for all m, since it must be a perfect power of p. Thus the polynomial
g takes the value pk infinitely often, so it must be identically equal to pk . (Otherwise the
polynomial g(x) pk would have infinitely many roots.) To finish the problem, note that
since g(x) = f (px) is constant, f (x) must be constant by the same logic.

B9. Let N be the set of non-negative integers. Suppose f : N N is a function such that
f (f (f (n))) < f (n + 1) for every n N . Prove that f (n) = n for all n using the following
steps or otherwise.
a) If f (n) = 0, then n = 0.

Let f (n) = 0. If n > 0, then n 1 is in the domain of f and f (f (f (n 1))) < f (n) = 0,
which is a contradiction, since 0 is the smallest possible value of f . (Note that this does
NOT prove that f (0) = 0, only that if f (some n) = 0, then that n = 0. In fact proving
f (0) = 0 along with part a would essentially solve the problem, see below.)

b) If f (x) < n, then x < n. (Start by considering n = 1.)

Induction on n. If n = 1, then this is just part a. Assuming the statement up to n we need

to prove that if f (x) < n + 1, then x < n + 1. If f (x) < n, then by induction x < n, so
x < n + 1. So let f (x) = n. If x = 0, we are done. Otherwise f (f (f (x 1))) < f (x) = n
and by using induction thrice we get in succession f (f (x 1)) < n, then f (x 1) < n and
then x 1 < n, i.e., x < n + 1 as desired.

c) f (n) < f (n + 1) and n < f (n + 1) for all n.

Apply part b to f (f (f (m))) < f (m + 1) (with x = f (f (m)) and n = f (m + 1)) to get

6
f (f (m)) < f (m + 1). Apply part b to this with x = f (m) and n = f (m + 1) to get
f (m) < f (m + 1). Again apply part b to get m < f (m + 1).

d) f (n) = n for all n.

By part c, f is increasing and f (n) n. If f (n) > n, then f (f (n)) > f (n) (since f
is increasing) and so f (f (n)) > n, i.e., f (f (n)) n + 1. Again, since f is increasing,
f (f (f (n))) f (n + 1), a contradiction.

Alternative solution after part a. Let us prove f (0) = 0. We know that f (n) = 0
implies n = 0, so n > 0 implies f (n) > 0. Applying this to any positive f (k), we get
f (f (k)) > 0. Denoting f (f (k)) = x, we therefore get f (f (f (x 1))) < f (x) = f (f (f (k))).
This means that for k such that f (f (f (k))) is the smallest number in {f (f (f (n)))|n 0},
we must have f (k) = 0. In particular 0 is in the range of f , so by part a f (0) = 0.
Since f (n) = 0 for no other n, we may restrict the function f by deleting 0 from the
domain and the range. The resulting function would satisfy f (f (f (n))) < f (n + 1) for
every n > 0. Repeat the reasoning substituting 1 (the new lowest element of the domain
and the range) for 0 and conclude f (1) = 1. Then restrict to n > 1 and show f (2) = 2
and so on.